vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

better way to disable for arm

Browse source
This commit is contained in:
slaren 2025-02-09 17:20:09 +01:00
parent 2d493d26ab
commit d4bdfc6314
1 changed files with 8 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

14
ggml/src/ggml-cpu/ggml-cpu.c
View file

@ -7824,21 +7824,23 @@ static void ggml_compute_forward_mul_mat_id(
const int64_t nr0 = ne01; const int64_t nr0 = ne01;
const int64_t nr1 = cne1; const int64_t nr1 = cne1;
#if defined(__aarch64__)
// disable for ARM
int chunk_size = (nr0 + nr1) / nth;
#else
int chunk_size = 16; int chunk_size = 16;
if (nr0 == 1 || nr1 == 1) { if (nr0 == 1 || nr1 == 1) {
chunk_size = 64; chunk_size = 64;
} }
#if defined(__aarch64__)
// disable for ARM
const bool disable_chunking = true;
#else
// disable for NUMA
const bool disable_chunking = ggml_is_numa();
#endif // defined(__aarch64__) #endif // defined(__aarch64__)
int64_t nchunk0 = (nr0 + chunk_size - 1) / chunk_size; int64_t nchunk0 = (nr0 + chunk_size - 1) / chunk_size;
int64_t nchunk1 = (nr1 + chunk_size - 1) / chunk_size; int64_t nchunk1 = (nr1 + chunk_size - 1) / chunk_size;
if (nchunk0 * nchunk1 < nth * 4 || ggml_is_numa()) { if (nchunk0 * nchunk1 < nth * 4 || disable_chunking) {
nchunk0 = nr0 > nr1 ? nth : 1; nchunk0 = nr0 > nr1 ? nth : 1;
nchunk1 = nr0 > nr1 ? 1 : nth; nchunk1 = nr0 > nr1 ? 1 : nth;
} }